Feedback Loops With Language Models Drive In-Context Reward Hacking.

Alexander Pan, Erik Jones, Meena Jagadeesan, Jacob Steinhardt

Browse the full ICML paper archive.